Buy-now-pay-later looks like the safest lending product ever invented. The tickets are tiny, the tenor is six weeks, the customer pays no interest, and the merchant foots the bill through a fee. So a lot of teams underwrite it as though it were barely credit at all — instant approval, minimal checks, scale as hard as the funnel allows. Then the loss line arrives, and per dollar lent it is worse than products ten times the ticket size. None of this is mysterious once you look at the actual unit economics, and, more importantly, the fix is almost never where teams instinctively look.

This is the long version: how BNPL really makes money, why that makes it so unforgiving, the full stack of loss drivers (two of which are structural and invisible), the single lever that moves loss the most, and the regulatory shift that is about to reprice the whole model. If you price, underwrite, or sit on the risk committee for a pay-later product, this is the mental model to carry into the room.

BNPL’s revenue comes mainly from the merchant discount rate, the merchant pays a few percent of the basket because the product lifts conversion and average order value. Late fees, and on longer plans some interest, are secondary. The point every credit person has to internalise is that revenue per loan is small and the tenor is short. A traditional installment loan earns interest over months or years; a pay-in-four plan earns a sliver of the basket, once. That single structural fact dictates how much credit risk you can actually afford to take, which is far less than the ‘it’s only six weeks’ intuition suggests.

Put the numbers in a single line and the fragility jumps out. Out of a few percent of gross revenue you pay credit losses, funding, and operating costs, and what survives is a sliver. Because the margin is so thin, a one-point move in the loss rate, the sort of move a personal-loan book would barely notice, can erase BNPL’s entire profit. The economics are unforgiving by design, not by accident, and that is the lens through which every other decision should be read.

Figure 1: Illustrative BNPL economics as a share of transaction value.

Revenue per loan is a sliver — so any single loss driver, left unmanaged, can wipe the entire margin.

It is never one thing. It is a stack of drivers that each look small and together are fatal to a thin margin. Losses are front-loaded, because instant, low-friction approval lets first-payment defaults and outright fraud through before you have earned a cent. On top of that sit two structural drivers most teams underestimate, plus an operational one that quietly bleeds the margin from the cost side.

Figure 2: Illustrative contribution of the main loss drivers.

The driver that surprises people most is loan stacking. A customer who looks responsible to you may be running five other pay-later plans across five other providers, and because BNPL has historically been thin or absent from credit bureaus, none of you can see the others. Each provider approves what looks like a modest, affordable purchase, and collectively they hand the customer far more short-term credit than anyone intended. This is not bad luck or weak underwriting on any single decision; it is a structural blind spot of the product. The fix is not a cleverer scorecard, it is visibility, which is exactly what bureau reporting (now arriving via regulation) provides.

The second structural trap is cohort drift. Your earliest cohorts look pristine because early adopters self-select, they are engaged, creditworthy, and careful. The board sees those numbers and extrapolates them into the growth plan. Then you scale: you broaden the customer base, raise approval rates, and push repeat frequency, and loss rates climb, often months after the rosy early figures were already baked into expectations. The discipline is to never read a young, narrow cohort as representative, and to watch loss by cohort vintage as you grow, not just the blended number.

Most BNPL teams underwrite the customer and stop there. But loss and fraud cluster hard by merchant and category, the same applicant is a very different risk buying electronics from a flash-sale site than buying groceries from an established retailer. Managing risk at the merchant level, pricing, limits, and monitoring by merchant and category, is the underused lever. Pair it with dynamic customer limits that expand only for proven repayers and choke off stackers fast, and start reporting to and pulling from the bureau so total leverage finally becomes visible. Those three moves attack the biggest leaks at once, and none of them is a marginal scorecard tweak.

The grey zone BNPL grew up in is closing. Affordability checks, clearer disclosures, late-fee limits, and bureau reporting are arriving market by market (the subject of a later issue). For economics this cuts both ways: fee caps and added underwriting cost squeeze an already-thin margin, but mandatory bureau reporting finally kills the stacking blind spot. Teams that treat the rules as a product feature, and use the new visibility to underwrite better, will end up with both a compliant and a more profitable book.

Stop staring at a single blended loss rate. Cut your losses by merchant category and by cohort vintage. That one view almost always reveals that a thin slice of merchants and your newest, broadest cohorts carry most of the loss, and unlike the headline number, that is something you can act on next week. It is the fastest path from ‘our losses are creeping up’ to ‘here is exactly where, and here is the fix.’

Next: building a scorecard with alternate & cash-flow data.

Views are my own and do not represent my employer.